Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation

Marica Muffoletto; Hao Xu; Hugo Barbaroux; Karl P. Kunze; Radhouene Neji; René Botnar; Claudia Prieto; Daniel Rueckert; Alistair Young

doi:10.1007/978-3-031-23443-9_9

Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation

Marica Muffoletto^*, Hao Xu, Hugo Barbaroux, Karl P. Kunze, Radhouene Neji, René Botnar, Claudia Prieto, Daniel Rueckert, Alistair Young

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

Abstract

Quantification of heart geometry is important in the clinical diagnosis of cardiovascular diseases. Changes in geometry are indicative of remodelling processes as the heart tissue adapts to disease. Coronary Computed Tomography Angiography (CCTA) is considered a first line tool for patients at low or intermediate risk of coronary artery disease, while Coronary Magnetic Resonance Angiography (CMRA) is a promising alternative due to the absence of radiation-induced risks and high performance in the evaluation of cardiac geometry. Yet, the accuracy of an image-based diagnosis is susceptible to the quality of volume segmentations. Deep Learning (DL) techniques are gradually being adopted to perform such segmentations and substitute the tedious and manual work performed by physicians. However, practical applications of DL techniques on a large scale are still limited due to their poor adaptability across modalities and patients. Hence, the aim of this work was to develop a pipeline to perform automatic heart segmentation of multiple cardiac imaging scans, addressing the domain shift between MRs (target) and CTs (source). We trained two Domain Adaptation (DA) methods, using Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs), following different training routines, which we refer to as un- and semi- supervised approaches. We also trained a baseline supervised model following state-of-the-art choice of parameters and augmentation. The results showed that DA methods can be significantly boosted by the addition of a few supervised cases, increasing Dice and Hausdorff distance metrics across the main cardiac structures.

Original language	English
Pages (from-to)	91-100
Number of pages	10
Journal	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
DOIs	https://doi.org/10.1007/978-3-031-23443-9_9
Publication status	Published - 2022
Event	13th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2022, held in conjunction with the 25th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2022 - Singapore, Singapore Duration: 18 Sept 2022 → 18 Sept 2022

Keywords

Automatic segmentation
Deep Learning
Domain adaptation
Generative Adversarial Networks
Variational Auto-Encoders

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1007/978-3-031-23443-9_9

Cite this

Muffoletto, M., Xu, H., Barbaroux, H., Kunze, K. P., Neji, R., Botnar, R., Prieto, C., Rueckert, D., & Young, A. (2022). Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 91-100. https://doi.org/10.1007/978-3-031-23443-9_9

@article{203daeacbc5e4cca809845878b85c2cf,

title = "Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation",

abstract = "Quantification of heart geometry is important in the clinical diagnosis of cardiovascular diseases. Changes in geometry are indicative of remodelling processes as the heart tissue adapts to disease. Coronary Computed Tomography Angiography (CCTA) is considered a first line tool for patients at low or intermediate risk of coronary artery disease, while Coronary Magnetic Resonance Angiography (CMRA) is a promising alternative due to the absence of radiation-induced risks and high performance in the evaluation of cardiac geometry. Yet, the accuracy of an image-based diagnosis is susceptible to the quality of volume segmentations. Deep Learning (DL) techniques are gradually being adopted to perform such segmentations and substitute the tedious and manual work performed by physicians. However, practical applications of DL techniques on a large scale are still limited due to their poor adaptability across modalities and patients. Hence, the aim of this work was to develop a pipeline to perform automatic heart segmentation of multiple cardiac imaging scans, addressing the domain shift between MRs (target) and CTs (source). We trained two Domain Adaptation (DA) methods, using Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs), following different training routines, which we refer to as un- and semi- supervised approaches. We also trained a baseline supervised model following state-of-the-art choice of parameters and augmentation. The results showed that DA methods can be significantly boosted by the addition of a few supervised cases, increasing Dice and Hausdorff distance metrics across the main cardiac structures.",

keywords = "Automatic segmentation, Deep Learning, Domain adaptation, Generative Adversarial Networks, Variational Auto-Encoders",

author = "Marica Muffoletto and Hao Xu and Hugo Barbaroux and Kunze, {Karl P.} and Radhouene Neji and Ren{\'e} Botnar and Claudia Prieto and Daniel Rueckert and Alistair Young",

note = "Funding Information: Research supported by ESPRC and Siemens Healthineers. Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 13th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2022, held in conjunction with the 25th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2022 ; Conference date: 18-09-2022 Through 18-09-2022",

year = "2022",

doi = "10.1007/978-3-031-23443-9_9",

language = "English",

pages = "91--100",

journal = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

issn = "0302-9743",

publisher = "Springer",

}

Muffoletto, M, Xu, H, Barbaroux, H, Kunze, KP, Neji, R , Botnar, R , Prieto, C, Rueckert, D & Young, A 2022, 'Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation', Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), pp. 91-100. https://doi.org/10.1007/978-3-031-23443-9_9

Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation. / Muffoletto, Marica; Xu, Hao; Barbaroux, Hugo et al.
In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2022, p. 91-100.

Research output: Contribution to journal › Article › peer-review

TY - JOUR

T1 - Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation

AU - Muffoletto, Marica

AU - Xu, Hao

AU - Barbaroux, Hugo

AU - Kunze, Karl P.

AU - Neji, Radhouene

AU - Botnar, René

AU - Prieto, Claudia

AU - Rueckert, Daniel

AU - Young, Alistair

PY - 2022

Y1 - 2022

N2 - Quantification of heart geometry is important in the clinical diagnosis of cardiovascular diseases. Changes in geometry are indicative of remodelling processes as the heart tissue adapts to disease. Coronary Computed Tomography Angiography (CCTA) is considered a first line tool for patients at low or intermediate risk of coronary artery disease, while Coronary Magnetic Resonance Angiography (CMRA) is a promising alternative due to the absence of radiation-induced risks and high performance in the evaluation of cardiac geometry. Yet, the accuracy of an image-based diagnosis is susceptible to the quality of volume segmentations. Deep Learning (DL) techniques are gradually being adopted to perform such segmentations and substitute the tedious and manual work performed by physicians. However, practical applications of DL techniques on a large scale are still limited due to their poor adaptability across modalities and patients. Hence, the aim of this work was to develop a pipeline to perform automatic heart segmentation of multiple cardiac imaging scans, addressing the domain shift between MRs (target) and CTs (source). We trained two Domain Adaptation (DA) methods, using Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs), following different training routines, which we refer to as un- and semi- supervised approaches. We also trained a baseline supervised model following state-of-the-art choice of parameters and augmentation. The results showed that DA methods can be significantly boosted by the addition of a few supervised cases, increasing Dice and Hausdorff distance metrics across the main cardiac structures.

AB - Quantification of heart geometry is important in the clinical diagnosis of cardiovascular diseases. Changes in geometry are indicative of remodelling processes as the heart tissue adapts to disease. Coronary Computed Tomography Angiography (CCTA) is considered a first line tool for patients at low or intermediate risk of coronary artery disease, while Coronary Magnetic Resonance Angiography (CMRA) is a promising alternative due to the absence of radiation-induced risks and high performance in the evaluation of cardiac geometry. Yet, the accuracy of an image-based diagnosis is susceptible to the quality of volume segmentations. Deep Learning (DL) techniques are gradually being adopted to perform such segmentations and substitute the tedious and manual work performed by physicians. However, practical applications of DL techniques on a large scale are still limited due to their poor adaptability across modalities and patients. Hence, the aim of this work was to develop a pipeline to perform automatic heart segmentation of multiple cardiac imaging scans, addressing the domain shift between MRs (target) and CTs (source). We trained two Domain Adaptation (DA) methods, using Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs), following different training routines, which we refer to as un- and semi- supervised approaches. We also trained a baseline supervised model following state-of-the-art choice of parameters and augmentation. The results showed that DA methods can be significantly boosted by the addition of a few supervised cases, increasing Dice and Hausdorff distance metrics across the main cardiac structures.

KW - Automatic segmentation

KW - Deep Learning

KW - Domain adaptation

KW - Generative Adversarial Networks

KW - Variational Auto-Encoders

UR - http://www.scopus.com/inward/record.url?scp=85147993702&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-23443-9_9

DO - 10.1007/978-3-031-23443-9_9

M3 - Article

AN - SCOPUS:85147993702

SN - 0302-9743

SP - 91

EP - 100

JO - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

JF - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

T2 - 13th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2022, held in conjunction with the 25th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2022

Y2 - 18 September 2022 through 18 September 2022

ER -